Fig. 1

Overview of the MetOncoFit Approach. a The MetOncoFit model consists of 142 metabolic features (Additional file 1: Table S1). These features include biochemical properties (e.g., catalytic activity (k_cat)), topological parameters from the RECON1 network model (e.g., biomass epicenter score), and dynamic properties computed from the NCI-60 cancer cell line metabolic models (e.g., reaction flux). These features for each gene are used to make predictions on its impact on tumor fitness in a specific cancer context. b The sample dataset in the figure shows the input matrix for MetOncoFit. The columns span the 142 features used in our model. The features are inputted into a random forest classification algorithm and it outputs ternary predictions (increased/neutral/decreased) for copy number variation, differential gene expression, and cancer patient survival based on the predicted impact of gene activity on tumor fitness. We evaluated MetOncoFit performance using 10-fold cross-validation. MetOncoFit also ranks features based on their predictive importance and outputs the direction of impact with expression, survival, or CNV (i.e., positive or negative correlation)